{"id":"arxiv/2602.03461","url":"https://arxiv.org/abs/2602.03461","tags":["cs.LG","math.OC","q-fin.CP","stat.ML"],"score":0,"title":"Soft-Radial Projection for Constrained End-to-End Learning","domain":"cs.LG","source":"arxiv","authors":["Philipp J. Schneider","Daniel Kuhn"],"reviews":0,"subject":"cs.LG","summary":"Integrating hard constraints into deep learning is essential for safety-critical systems. Yet existing constructive layers that project predictions onto constraint boundaries face a fundamental bottleneck: gradient saturation. By collapsing exterior points onto lower-dimensional surfaces, standard orthogonal projections induce rank-deficient Jacobians, which nullify gradients orthogonal to active constraints and hinder optimization. We introduce Soft-Radial Projection, a differentiable reparameterization layer that circumvents this issue through a radial mapping from Euclidean space into the…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-02-03","reactions":{}}